Use of Machine Learning Methods to Reduce Predictive Error of Groundwater Models

Quantitative analyses of groundwater flow and transport typically rely on a physically‐based model, which is inherently subject to error. Errors in model structure, parameter and data lead to both random and systematic error even in the output of a calibrated model. We develop complementary data‐driven models (DDMs) to reduce the predictive error of physically‐based groundwater models. Two machine learning techniques, the instance‐based weighting and support vector regression, are used to build the DDMs. This approach is illustrated using two real‐world case studies of the Republican River Compact Administration model and the Spokane Valley‐Rathdrum Prairie model. The two groundwater models have different hydrogeologic settings, parameterization, and calibration methods. In the first case study, cluster analysis is introduced for data preprocessing to make the DDMs more robust and computationally efficient. The DDMs reduce the root‐mean‐square error (RMSE) of the temporal, spatial, and spatiotemporal prediction of piezometric head of the groundwater model by 82%, 60%, and 48%, respectively. In the second case study, the DDMs reduce the RMSE of the temporal prediction of piezometric head of the groundwater model by 77%. It is further demonstrated that the effectiveness of the DDMs depends on the existence and extent of the structure in the error of the physically‐based model.     

Forward induced seismic hazard assessment: application to a synthetic seismicity catalogue from hydraulic stimulation modelling

The M _w 3.2-induced seismic event in 2006 due to fluid injection at the Basel geothermal site in Switzerland was the starting point for an ongoing discussion in Europe on the potential risk of hydraulic stimulation in general. In particular, further development of mitigation strategies of induced seismic events of economic concern became a hot topic in geosciences and geoengineering. Here, we present a workflow to assess the hazard of induced seismicity in terms of occurrence rate of induced seismic events. The workflow is called Forward Induced Seismic Hazard Assessment (FISHA) as it combines the results of forward hydromechanical-numerical models with methods of time-dependent probabilistic seismic hazard assessment. To exemplify FISHA, we use simulations of four different fluid injection types with various injection parameters, i.e. injection rate, duration and style of injection. The hydromechanical-numerical model applied in this study represents a geothermal reservoir with preexisting fractures where a routine of viscous fluid flow in porous media is implemented from which flow and pressure driven failures of rock matrix and preexisting fractures are simulated, and corresponding seismic moment magnitudes are computed. The resulting synthetic catalogues of induced seismicity, including event location, occurrence time and magnitude, are used to calibrate the magnitude completeness M _c and the parameters a and b of the frequency-magnitude relation. These are used to estimate the time-dependent occurrence rate of induced seismic events for each fluid injection scenario. In contrast to other mitigation strategies that rely on real-time data or already obtained catalogues, we can perform various synthetic experiments with the same initial conditions. Thus, the advantage of FISHA is that it can quantify hazard from numerical experiments and recommend a priori a stimulation type that lowers the occurrence rate of induced seismic events. The FISHA workflow is rather general and not limited to the hydromechanical-numerical model used in this study and can therefore be applied to other fluid injection models.     

Wave data assimilation using a hybrid approach in the Persian Gulf

The main goal of this study is to develop an efficient approach for the assimilation of the hindcasted wave parameters in the Persian Gulf. Hence, the third generation SWAN model was employed for wave modeling forced by the 6-h ECMWF wind data with a resolution of 0.5°. In situ wave measurements at two stations were utilized to evaluate the assimilation approaches. It was found that since the model errors are not the same for wave height and period, adaptation of model parameter does not result in simultaneous and comprehensive improvement of them. Therefore, an approach based on the error prediction and updating of output variables was employed to modify wave height and period. In this approach, artificial neural networks (ANNs) were used to estimate the deviations between the simulated and measured wave parameters. The results showed that updating of output variables leads to significant improvement in a wide range of the predicted wave characteristics. It was revealed that the best input parameters for error prediction networks are mean wind speed, mean wind direction, wind duration, and the wave parameters. In addition, combination of the ANN estimated error with numerically modeled wave parameters leads to further improvement in the predicted wave parameters in contrast to direct estimation of the parameters by ANN.     

Estimation of enterococci input from bathers and animals on a recreational beach using camera images

Enterococci, are used nationwide as a water quality indicator of marine recreational beaches. Prior research has demonstrated that enterococci inputs to the study beach site (located in Miami, FL) are dominated by non-point sources (including humans and animals). We have estimated their respective source functions by developing a counting methodology for individuals to better understand their non-point source load impacts. The method utilizes camera images of the beach taken at regular time intervals to determine the number of people and animal visitors. The developed method translates raw image counts for week days and weekend days into daily and monthly visitation rates. Enterococci source functions were computed from the observed number of unique individuals for average days of each month of the year, and from average load contributions for humans and for animals. Results indicate that dogs represent the larger source of enterococci relative to humans and birds.     

Earthquake Stability Analysis of Rock Slopes: a Case Study

Stability analysis of Surabhi landslide in the Dehradun and Tehri districts of Uttaranchal located in Mussoorie, India, has been simulated numerically using the distinct element method focusing on the weak zones (fracture). This is an active landslide on the main road toward the town centre, which was triggered after rainfall in July–August 1998. Understanding the behaviour of this landslide will be helpful for planning and implementing mitigation measures. The first stage of the study includes the total area of the landslide. The area identified as the zone of detachment is considered the most vulnerable part of the landslide. Ingress of water and increased pore pressures result in reduced mobilized effective frictional resistance, causing the top layer of the zone of detachment to start moving. The corresponding total volume of rock mass that is potentially unstable is estimated to 11.58 million m³. The second stage of this study includes a 2D model focussing only on the zone of detachment. The result of the analyses including both static and dynamic loading indicates that most of the total displacement observed in the slide model is due to the zone of detachment. The discontinuum modelling in the present study gives reasonable agreement with actual observations and has improved understanding of the stability of the slide slope.     

A hybrid computational approach to formulate soil deformation moduli obtained from PLT

In this study, new empirical equations were developed to predict the soil deformation moduli utilizing a hybrid method coupling genetic programming and simulated annealing, called GP/SA. The proposed models relate secant (E_s), unloading (E_u) and reloading (E_r) moduli obtained from plate load–settlement curves to the basic soil physical properties. Several models with different combinations of the influencing parameters were developed and checked to select the best GP/SA models. The database used for developing the models was established upon a series of plate load tests (PLT) conducted on different soil types at various depths. The validity of the models was tested using parts of the test results that were not included in the analysis. The validation of the models was further verified using several statistical criteria. A traditional GP analysis was performed to benchmark the GP/SA models. The contributions of the parameters affecting E_s, E_u and E_r were analyzed through a sensitivity analysis. The proposed models are able to estimate the soil deformation moduli with an acceptable degree of accuracy. The E_s prediction model has a remarkably better performance than the models developed for predicting E_u and E_r. The simplified formulations for E_s, E_u and E_r provide significantly better results than the GP-based models and empirical models found in the literature.     

The use of geomorphological elements for evaluation of visual quality of Israeli coast

This study describes a method for quantitative evaluation of visual resources of the Mediterranean coast of Israel. The method used visual characteristics of contrast, variety, competition, composition and harmony to identify visually preferred sites. Various coastal physical elements were used as a measurement scale of the characteristics. The study also examined the use of photographic slides vs. filed visit as a tool for visual resources evaluation of the Israeli coast. Findings were based on data gathered from a questionnaire given to two groups in the laboratory and in the field. Statistical analysis of the findings indicated that the proposed method could be used effectively to measure visual preference of the Israeli coast using physical elements and quantitative values. The data also show that the use of photographic slides in laboratory conditions is as effective a tool as field observation for the evaluation of observer's visual preference of the coast.     

Lithology-dependent seismic anisotropic amplitude variation with offset correction in transversely isotropic media

Analysis of amplitude variation with offset is an essential step for reservoir characterization. For an accurate reservoir characterization, the amplitude obtained with an isotropic assumption of the reservoir must be corrected for the anisotropic effects. The objective is seismic anisotropic amplitude correction in an effective medium, and, to this end, values and signs of anisotropic parameter differences (Δδ and Δε) across the reflection interfaces are needed. These parameters can be identified by seismic and well log data. A new technique for anisotropic amplitude correction was developed to modify amplitude changes in seismic data in transversely isotropic media with a vertical axis of symmetry. The results show that characteristics of pre-stack seismic data, that is, amplitude variation with offset gradient, can be potentially related to the sign of anisotropic parameter differences (Δδ and Δε) between two layers of the reflection boundary. The proposed methodology is designed to attain a proper fit between modelled and observed amplitude variation with offset responses, after anisotropic correction, for all possible lithofacies at the reservoir boundary. We first estimate anisotropic parameters, that is, δ and ε, away from the wells through Backus averaging of elastic properties resulted from the first pass of isotropic pre-stack seismic inversion, on input data with no amplitude correction. Next, we estimate the anisotropic parameter differences at reflection interfaces (values and signs of Δδ and Δε). We then generate seismic angle gather data after anisotropic amplitude correction using Rüger's equation for the P-P reflection coefficient. The second pass of isotropic pre-stack seismic inversion is then performed on the amplitude-corrected data, and elastic properties are estimated. Final outcome demonstrates how introduced methodology helps to reduce the uncertainty of elastic property prediction. Pre-stack seismic inversion on amplitude-corrected seismic data results in more accurate elastic property prediction than what can be obtained from non-corrected data. Moreover, a new anisotropy attribute (ν) is presented for improvement of lithology identification.     

Forecasting flood-prone areas using Shannon’s entropy model

With regard to the lack of quality information and data in watersheds, it is of high importance to present a new method for evaluating flood potential. Shannon’s entropy model is a new model in evaluating dangers and it has not yet been used to evaluate flood potential. Therefore, being a new model in determining flood potential, it requires evaluation and investigation in different regions and this study is going to deal with this issue. For to this purpose, 70 flooding areas were recognized and their distribution map was provided by ArcGIS10.2 software in the study area. Information layers of altitude, slope angle, slope aspect, plan curvature, drainage density, distance from the river, topographic wetness index (TWI), lithology, soil type, and land use were recognized as factors affecting flooding and the mentioned maps were provided and digitized by GIS environment. Then, flood susceptibility forecasting map was provided and model accuracy evaluation was conducted using ROC curve and 30% flooding areas express good precision of the model (73.5%) for the study area.